Abstract
In this paper, the geodesic distance is applied to relational clustering methods. First, it is shown that conventional methods are based on respective three types of relational clustering algorithms among nine ones, and the six rests of the nine ones with the geodesic distance are proposed. Second, geodesic dissimilarity is proposed by assigning the power of the Euclidean distance to the weight of the neighborhood graph of data. Numerical examples show that the proposed geodesic-dissimilarity-based relational clustering algorithms successfully cluster the data that conventional squared-Euclidean-distance-based ones cannot.
